We propose to use computational modeling and neuroimaging of the Stroop task to investigate the neural and computational basis of selective attention. In this task, participants must selectively attend and respond to the font color of color words while ignoring the words themselves. Most previous imaging studies of Stroop have investigated the anatomical source of selective attention and less attention has been paid to the behavior of the posterior targets of that top-down control. We propose to use functional magnetic resonance imaging (fMRI) to study the posterior targets specifically and to extend previous computational models of Stroop to address our findings. Our specific aims are to (1) use fMRI to investigate whether top-down control in the Stroop task excites relevant information, inhibits irrelevant information, or both, (2) use fMRI to investigate the way in which subject strategies influence word processing during a semantic associate variant of the Stroop task, and (3) extend previous connectionist models of the Stroop task to address our empirical findings. These studies on the Stroop task will potentially make important contributions to our understanding of the neural and computational underpinnings of selective attention more generally.